/DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the Amendment filed on 8/26/2026.
Claims 1, 4-16, 19-24 are pending. Claims 1, 4-11, 13-14, 19-20 have been amended. Claims 2-3, 17-18 are cancelled. Claims 21-24 are newly added.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-7, 13-17, 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Truong et al. (US 10460235 B1, hereinafter Truong), in view of Raman et al. (US 11032063 B2, hereinafter Raman) further in view of Kalchbrenner et al. (US 11468295 B2, hereinafter Kalchbrenner).
Regarding Claim 1, Truong teaches a data generation apparatus comprising: at least one processor; and at least one memory storing at least one instruction that, when executed by the at least one processor, causes the at least one processor to (Truong, Column 7, line 14-18, "The system for generating data models can include at least one secure system processor and at least one secure system non-transitory memory"; Column 4, Line 11-15, "The cloud computing system can include at least one processor and at least one non-transitory memory storing instructions that, when executed by the at least one processor cause the cloud computing system to perform the following operations"): [[generate latent information to be input into a generative model by utilizing latent information that is represented in a blockchain compliant code,]] generate data by inputting the generated latent information into the generative model (Truong, Column 27, Line 65-66, "a decoder maps from a code space <read on generated latent information> to a sample space . ... generating a sample <read on generating data> using the decoder"; Column 10, line 13-15, "the data model can be configured to map from a random or pseudorandom vector <read on generated latent information> to elements in the training data space"; Column 7, Line 33-35, "generating synthetic data using the data model in response to the data generation request"), wherein the generative model includes a neural network (Truong, Column 11, line 48-50, "neural network, recurrent neural network, generative adversarial network <read on generative model>"; Column 2, Line 14-16, "a dataset generator can generate a synthetic dataset for training the data model using a generative network of a generative adversarial network"), and [[wherein the generated data includes at least one of an image, a video, or a sound.]]
But Truong does not explicitly disclose generate latent information to be input into a generative model by utilizing latent information that is represented in a blockchain compliant code; and wherein the generated data includes at least one of an image, a video, or a sound.
However, Raman teaches generate latent information to be input into a generative model by utilizing latent information that is represented in a blockchain compliant code (Raman, Column 9, Line 49-50, "Once validated by a subset, the validated frame of data <read on latent information that is represented in a blockchain compliant code> may be stored on the blockchain 160"; Column 18, line 17-19, "the data frame <read on latent information that is represented in a blockchain compliant code> may be decompressed to reveal the state of the simulation <read on generated latent information to be input into a generative model> before the compression").
Raman and Truong are analogous since both are dealing with the management, storage, and transformation of high-dimensional vectors representing model states or latent information in distributed or cloud computing environments. Truong provided a way of generating synthetic data by mapping from a code space (latent) representation to a sample space using a generative decoder. Raman provided a way of storing compressed state vectors as validated frames on a blockchain ledger and then decompressing those frames to reveal the underlying state vector for further downstream processing. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the blockchain-based frame storage and decompression scheme taught by Raman into the data generation system of Truong such that the latent information driving the generative model is represented and stored as a blockchain-compliant code and is decompressed to produce the latent information that is then input into the generative model. The motivation is to provide a shared, immutable, append-only record so that entities in the system can trust the validity of the conclusions obtained by individual agents, as discussed by Raman.
Kalchbrenner teaches wherein the generated data includes at least one of an image, a video, or a sound (Kalchbrenner, Column 3, line 14-24, "output example 152 can be a sequence of audio data <read on sound>, e.g., a waveform . ... As another example, the output example 152 can be an image . ... As another example, the output example 152 can be a video frame <read on video>").
Kalchbrenner and Truong are analogous since both are dealing with the generation of synthetic approximations of real-world data using neural network models. Truong provided a way of creating synthetic datasets that mimic the structure and statistics of underlying data by mapping from latent vectors to output samples through a generative network. Kalchbrenner provided a way of generating high-quality output examples including audio waveforms, images, and video frames from a neural network model. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the multimedia output generation capabilities taught by Kalchbrenner into the modified invention of Truong and Raman such that the data generated by the generative model includes at least one of an image, a video, or a sound. The motivation is to leverage the structure of output examples to achieve speed ups without degradation in quality, as discussed by Kalchbrenner.
Regarding Claim 2, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the at least one processor is configured to generate the data by inputting the latent information into the generative model (Truong, Column 27, Line 65-66, "a decoder maps from a code space <read on latent information> to a sample space. ... generating a sample using the decoder").
Regarding Claim 3, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches [[ wherein the at least one processor is configured to convert the latent information that is represented in the blockchain compliant code to latent information that is not represented in the blockchain compliant code, ]] generate the data by inputting the converted latent information into the generative model (Truong, Column 2, Line 63-67, "The generative network can include a decoder network configured to generate decoder output data in a sample space having a first dimensionality from decoder input data in a code space having a second dimensionality less than the first dimensionality").
Truong does not explicitly disclose but Raman teaches wherein the at least one processor is configured to convert the latent information that is represented in the blockchain compliant code to latent information that is not represented in the blockchain compliant code (Raman, Column 18, Line 17-19, "the data frame may be decompressed to reveal the state of the simulation before the compression"; it is noted the "state" represents the underlying variables needed to continue iterative computation).
Raman and Truong are analogous since both of them are dealing with the management and generation of simulation states and synthetic data models using distributed computing and blockchain ledgers. Truong provided a way of generating synthetic datasets by mapping from a "code space" (latent information) to a "sample space" using a generative decoder. Raman provided a way of storing simulation state vectors on a blockchain and then decompressing them to reveal the original state (converting from blockchain code to non-compliant information) for further processing. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the decompression and state-reconstruction mechanisms taught by Raman into the modified invention of Truong such that blockchain-stored latent variables can be converted and input into the generative model. The motivation is to ensure the results inferred by individual agents are trustworthy and transferable.
Regarding Claim 4, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the at least one instruction causes the at least one processor to convert the latent information that is represented in the blockchain compliant code to the generated latent information based on predetermined correspondence (Truong, Column 1, line 64-66, "tokenization can result in tokenized data, sensitive data values, and a mapping <read on predetermined correspondence> between the tokens and the values"; Column 2, Line 63-67, "The generative network can include a decoder network configured to generate decoder output data in a sample space having a first dimensionality from decoder input data in a code space having a second dimensionality less than the first dimensionality").
Truong does not explicitly disclose but Raman teaches wherein the at least one instruction causes the at least one processor to convert the latent information that is represented in the blockchain compliant code to the generated latent information (Raman, Column 18, line 17-19, "the data frame <read on latent information that is represented in the blockchain compliant code> may be decompressed to reveal the state of the simulation <read on generated latent information> before the compression"), based on predetermined correspondence (Raman, Column 8, line 61-62, "constructs frames of states of the simulation and performs delta encoding and lattice vector quantization <read on predetermined correspondence> to compress the frames").
Raman and Truong are analogous since both are dealing with the transformation of high-dimensional vectors between compact and expanded representations. Truong provided a way of protecting and mapping data values through tokenization with a defined correspondence between tokens and underlying values. Raman provided a way of translating between raw simulation state vectors and blockchain-compliant compressed frames using a mathematical schema based on delta encoding and lattice vector quantization. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the blockchain storage and decompression mechanism taught by Raman into the data generation system of Truong such that latent information stored in the efficient blockchain-compliant format is converted back to usable latent information for input to the generative model using the predetermined mathematical correspondence. The motivation is to ensure the integrity, security, and storage efficiency of the generation parameters.
Regarding Claim 5, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 4.
The combination further teaches wherein the predetermined correspondence is between an element of the latent information that is represented in the blockchain compliant code and an element of the generated latent information (Truong, Column 1, line 64-67, "tokenization can result in tokenized data, sensitive data values, and a mapping <read on correspondence> between the tokens <read on element of the latent information represented in the blockchain compliant code> and the values <read on element of the generated latent information>").
Regarding Claim 6, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the latent information that is represented in the blockchain compliant code has a smaller data amount than the generated latent information (Raman, Column 20, line 20-23, "compressing the simulation content within the data frame based on previous simulation content stored in another data frame to generate a compressed data frame <read on smaller data amount blockchain compliant code>").
Raman and Truong are analogous since both are dealing with the efficient storage and transmission of high-dimensional vector data in a distributed network. Truong provided a way of defining a reduced-dimensionality code space to represent complex data features for a generative model. Raman provided a way of reducing network traffic demand by performing delta encoding and lattice vector quantization to generate compressed data frames for blockchain storage. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the compression and quantization scheme taught by Raman into the modified invention of Truong such that the latent information represented in the blockchain compliant code occupies a smaller data amount than the generated latent information that is input into the generative model. The motivation is to reduce network traffic demand and reduce hardware resources.
Regarding Claim 7, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the generated latent information includes information for characterizing an object (Truong, Column 17, line 5-6, "classes could include employee identification numbers, employee names, employee addresses <read on characterizing an employee object>").
Regarding Claim 13, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches [[wherein the at least one instruction causes the at least one processor to generate the latent information by performing at least one of an arithmetic operation or a logical operation]] using the first latent information and the second latent information (Truong, Column 3, Line 6-10, "Generating the synthetic dataset ... can further include determining a first representative point ... and a second representative point ... and determining a vector connecting the first representative point and the second representative point <read on using the first latent information and the second latent information>").
Truong does not explicitly disclose wherein the at least one instruction causes the at least one processor to generate the latent information by performing at least one of an arithmetic operation or a logical operation.
However, Raman teaches wherein the at least one instruction causes the at least one processor to generate the latent information by performing at least one of an arithmetic operation or a logical operation (Raman, Column 8, Line 60-65, "performs delta encoding <read on arithmetic operation> and lattice vector quantization to compress the frames"), using the first latent information and the second latent information (Raman, Column 21, line 25-27, "iterative simulation content may be compressed based on differences in state with a previous iteration <read on first and second latent information>").
Raman and Truong are analogous since both are dealing with the mathematical manipulation of high-dimensional vectors representing model states or latent information. Truong provided a way of finding a difference vector between two representative points in a code space to enable the translation and generation of other points. Raman provided a way of generating compressed state updates by performing delta encoding, which involves calculating arithmetic differences between a current state and a previous state. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the arithmetic delta encoding techniques taught by Raman into the latent information generation process of Truong such that new latent codes are derived from the arithmetic combination of existing latent points. The motivation is to provide a mathematically efficient way to represent and evolve the underlying parameters of a generative model.
Regarding Claim 14, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 10.
The combination further teaches wherein the first latent information is the latent information that is represented in the blockchain compliant code and the second latent information is represented in the blockchain compliant code (Raman, Column 9, Line 49-50, "Once validated by a subset, the validated frame of data <read on latent information represented in the blockchain compliant code> may be stored on the blockchain 160"; Column 10, line 1-10, "system state vector θt ∈ d' is a shared source of (possibly random) information ... report the state Xt at checkpoints t ∈{T1, T2, ...} .... report compressed update t").
Raman and Truong are analogous since both are dealing with the secure storage and validation of multiple state vectors and latent parameters in a distributed system. Truong provided a way of determining first and second representative points in a code space to enable the generation or interpolation of synthetic data. Raman provided a way of committing ordered frame data of multiple evolving state vectors to a distributed ledger shared among multiple nodes. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the blockchain-based ledger storage taught by Raman into the modified invention of Truong such that both the first latent information and the second latent information used to derive the input to the generative model are represented in a blockchain compliant code. The motivation is to ensure that entities in the system can trust the validity of the conclusions obtained by individual agents.
Regarding Claim 15, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 10.
The combination further teaches wherein the first latent information is purchased from a first user by a second user (Truong, Column 16, Line 60-65, "The actual data may have been purchased in whole or in part by an entity associated with system 100").
Regarding Claim 16, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 10.
The combination further teaches wherein the first latent information includes information for characterizing a first object, and wherein the second latent information includes information for characterizing a second object (Truong, Column 28, line 29-31, "identify one or more first accounts belonging to users in their 20s <read on first object> and one or more second accounts belonging to users in their 40s <read on second object>").
Regarding Claim 19, it recites limitations similar in scope to the limitations of Claim 1 but as a method and the combination of Truong, Raman, and Kalchbrenner teaches all the limitations as of Claim 1. Therefore is rejected under the same rationale.
Regarding Claim 20, it recites limitations similar in scope to the limitations of claim 1 and the combination of Truong, Raman, and Kalchbrenner teaches all the limitations as of Claim 1. And Truong discloses these features can be implemented on a computer-readable storage medium (Truong, Column 38, Line 6-10, “Furthermore, although aspects of the disclosed embodiments are described as being associated with data stored in memory and other tangible computer-readable storage mediums”; Column 4, Line 11-15, “The cloud computing system can include at least one processor and at least one non-transitory memory storing instructions that, when executed by the at least one processor cause the cloud computing system to perform the following operations”).
Regarding Claim 22, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the generated latent information includes noise information (Truong, Column 10, line 13-15, "the data model can be configured to map from a random or pseudorandom vector <read on noise information> to elements in the training data space").
Regarding Claim 23, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the generated latent information is latent information that is not represented in the blockchain compliant code (Raman, Column 18, line 17-19, "the data frame may be decompressed to reveal the state of the simulation <read on latent information that is not represented in the blockchain compliant code> before the compression").
Raman and Truong are analogous as discussed in Claim 1 above. The obviousness rationale for combining Raman's decompression scheme into Truong's data generation system to produce a decompressed latent representation for input to the generative model is provided above and is incorporated here.
Regarding `, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination further teaches wherein the latent information to be input into the generative model is generated using a random component (Truong, Column 10, line 13-15, "the data model can be configured to map from a random or pseudorandom vector <read on random component> to elements in the training data space").
Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over Truong et al. (US 10460235 B1, hereinafter Truong), in view of Raman et al. (US 11032063 B2, hereinafter Raman) as applied to Claim 1 above and further in view of Liu et al. (US 20180247201 A1, hereinafter Liu).
Regarding Claim 8, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination does not explicitly disclose but Liu teaches wherein the generated latent information includes information on at least one of a hair style, a hair color, an eye color, a skin color, an expression, or an attachment of the object (Liu, Paragraph [0061], "Examples of face attributes, include hair color, expression, facial hair, and eyeglasses <read on attachment>").
Liu and Truong are analogous since both are dealing with the classification and generation of objects using machine-learned feature vectors and attributes. Truong provided a way of identifying sensitive data classes to generate specific synthetic portions for objects like employees or financial accounts. Liu provided a way of translating and generating specific face attributes including hair color, expression, and eyeglasses within a latent space. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the specific visual attribute types taught by Liu into the modified invention of Truong, Raman, and Kalchbrenner such that the generated latent information includes specific descriptive characteristics of an object like hair style or expression. The motivation is to accurately differentiate between target classes and capture biological variance.
Claim(s) 9, 10, 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Truong et al. (US 10460235 B1, hereinafter Truong), in view of Raman et al. (US 11032063 B2, hereinafter Raman) as applied to Claim 1 above and further in view of Srinivasan (US 20190266442 A1).
Regarding Claim 9, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination does not explicitly disclose but Srinivasan teaches wherein the generated latent information includes a characteristic and a variable characteristic of the object (Srinivasan, Paragraph [0045], "first image and a second input, which is indicative of a desire for more or less of the at least one user-defined attribute <read on variable characteristic>").
Srinivasan and Truong are analogous since both are dealing with the refinement of generative model outputs based on specific user-defined parameters. Truong provided a way of generating synthetic data using class-specific and subclass-specific models to match the statistical distributions of real data. Srinivasan provided a way of generating an image with a visual attribute (characteristic) and then tuning that attribute with a second input indicative of "more or less" of it (variable characteristic). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the tunable attribute stacking taught by Srinivasan into the modified invention of Truong, Raman, and Kalchbrenner such that the generated latent information includes both a base characteristic and a relative variable characteristic. The motivation is to enable the end-user to vary one or more visual attributes of an item in an image in a relative manner.
Regarding Claim 10, the combination of Truong, Raman, and Kalchbrenner teaches the invention in Claim 1.
The combination does not explicitly disclose but Srinivasan teaches wherein the at least one instruction causes the at least one processor to generate the latent information based on a first latent information and a second latent information (Srinivasan, Paragraph [0013], "a tunable adversarial network includes a stacked GAN network, wherein a first GAN ... generate an image of an item with a visual attribute <read on first latent information> and a second GAN ... generate an image of the item with more or less of the visual attribute <read on second latent information>").
Srinivasan and Truong are analogous since both are dealing with the derivation of new data representations by combining or evolving multiple input sources. Truong provided a way of determining a vector connecting a first representative point and a second representative point in code space to translate or generate new data. Srinivasan provided a way of using a stacked GAN configuration where a first GAN's output and a second user input are used together to generate a final result. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the stacked generation architecture taught by Srinivasan into the modified invention of Truong, Raman, and Kalchbrenner such that new latent information is generated based on a combination of first and second latent information inputs. The motivation is to improve the accuracy of images generated via a GAN.
Regarding Claim 21, the combination of Truong, Raman, Kalchbrenner, and Srinivasan teaches the invention in Claim 10.
The combination further teaches wherein the first latent information is not represented in the blockchain compliant code and the second latent information is not represented in the blockchain compliant code (Truong, Column 10, line 13-15, "the data model can be configured to map from a random or pseudorandom vector <read on latent information not represented in the blockchain compliant code> to elements in the training data space"; Column 3, Line 6-10, "determining a first representative point ... and a second representative point <read on first and second latent information not represented in the blockchain compliant code>").
Claim(s) 11, 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Truong et al. (US 10460235 B1, hereinafter Truong), in view of Raman et al. (US 11032063 B2, hereinafter Raman), further in view of Srinivasan (US 20190266442 A1) as applied to Claim 10 above and further in view of Aliper et al. (US 20190034581 A1, hereinafter Aliper).
Regarding Claim 11, the combination of Truong, Raman, Kalchbrenner, and Srinivasan teaches the invention in Claim 10.
The combination does not explicitly disclose but Aliper teaches wherein the at least one instruction causes the at least one processor to generate the latent information by performing a genetic operation using the first latent information and the second latent information (Aliper, Paragraph [0267], "a specific type of DNN called Deep Feature Selection (DFS) is trained on blood gene expression samples using standard backpropagation algorithm ... the DFS model is applied to select a set of age-related genes using different DNN-based feature selection methods combined into one ensemble model via genetic algorithm"; Paragraph [0281], "On each iteration the GA performed 35 crossover operations between its populations and 15 mutation operations, during which random genes were injected in the training of GA").
Aliper and Truong are analogous since both are dealing with the generation and optimization of data representations using machine learning models in computational data processing environments. Truong provided a way of generating synthetic data using generative models that map from latent vectors to elements in a training data space. Aliper provided a way of using a genetic algorithm to combine and evolve multiple latent feature representations through crossover and mutation operations to produce optimized new latent information. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to incorporate the genetic algorithm-based combination of latent feature representations taught by Aliper into the modified invention of Truong, Raman, Kalchbrenner, and Srinivasan such that the latent information is generated by performing a genetic operation using the first latent information and the second latent information. The motivation is to obtain an optimized final ranking list that yields improved prediction accuracy, as discussed by Aliper in Paragraph [0281].
Regarding Claim 12, the combination of Truong, Raman, Srinivasan, and Aliper teaches the invention in Claim 11.
Th combination further teaches wherein the genetic operation includes at least one of crossover, mutation, or selection (Aliper, Paragraph [0281], " by applying the forementioned feature selection algorithms on both DNN and DFS models. On each iteration the GA performed 35 crossover operations between its populations and 15 mutation operations, during which random genes were injected in the training of GA").
As explained in rejection of claim 11, the obviousness for combining of genetic algorithm of Aliper into Truong is provided above.
Response to Arguments
The rejection of Claims 1-14, 16-20 under Nonstatutory Double Patenting are withdrawn in view of Applicant filed eTerminal/Terminal Disclaimer on 8/26/2026.
Applicant’s arguments with respect to claim 1, 19, 20, filed on 8/26/2026, with respect to rejection under 35 USC § 103 have been considered but is not persuasive.
Applicant asserts that Raman does not disclose "latent information that is represented in a blockchain compliant code" because Raman's frame includes compressed state updates and state data within a previous frame (i.e., the system state vector Xt and its difference), and does not disclose that the frame includes the shared source of information Θt.
In response to the argument, Examiner respectfully disagrees. The claim limitation only requires that "latent information" be "represented in a blockchain compliant code" — it does not require that the latent information specifically correspond to Raman's Θt. As described in the rejection of Claim 1 above, Raman in Column 9, Lines 49-50 expressly teaches "Once validated by a subset, the validated frame of data may be stored on the blockchain 160," and Raman in Column 10, Lines 1-10 further teaches that the system state vector, together with a shared source of (possibly random) information, drives the iterative evolution of the simulation. The system state vector Xt is itself high-dimensional data that characterizes the underlying model at each checkpoint and reads on "latent information" under the broadest reasonable interpretation, consistent with Applicant's own specification (S0.txt, Paragraph [0037]) which defines latent information as "information that may be put into a latent variable inferred from observation data ... through a model or the like." Because the validated frame (containing Xt-derived compressed state data) is expressly stored on the blockchain 160, Raman does teach latent information that is represented in a blockchain compliant code. Whether Θt is or is not stored on the blockchain is not dispositive of the rejection.
Applicant further asserts that Truong and Raman are silent about "the flow of generating latent information by utilizing latent information that is represented in a blockchain compliant code, and generating data by inputting the generated latent information into the generative model."
In response to the argument, Examiner respectfully disagrees. As described in the rejection of Claim 1 above, this two-step flow is taught by the combination of Truong and Raman. In particular, Raman in Column 18, Lines 17-19 teaches that "the data frame may be decompressed to reveal the state of the simulation before the compression" — that is, the blockchain-stored data frame (latent information represented in a blockchain compliant code) is decompressed to produce a distinct, decoded state representation (generated latent information) that is subsequently usable for downstream computation. Truong in Column 27, Lines 65-66 then teaches that "a decoder maps from a code space to a sample space . ... generating a sample using the decoder," and in Column 10, Lines 13-15 that "the data model can be configured to map from a random or pseudorandom vector to elements in the training data space." The ordered combination — decompress the blockchain-stored latent (Raman) → obtain the decoded latent representation → input the decoded latent representation into the generative model to produce output data (Truong) — reads directly on the claimed two-step flow of "generate latent information to be input into a generative model by utilizing latent information that is represented in a blockchain compliant code, and generate data by inputting the generated latent information into the generative model." The motivation to combine, as set forth in the rejection of Claim 1 above, is to provide a shared, immutable, append-only record so that the parameters driving synthetic data generation are trustworthy and transferable, as discussed by Raman. Hence the combination of Truong and Raman fully teaches the disputed limitation. Therefore Applicant's remark cannot be considered persuasive.
Applicant further asserts that the amended claim 1 requires "wherein the generative model includes a neural network, and wherein the generated data includes at least one of an image, a video, or a sound," and that neither Truong nor Raman teaches this feature.
In response to the argument, Examiner respectfully disagrees. As described in the rejection of Claim 1 above, Truong in Column 11, Lines 48-50 expressly teaches "neural network, recurrent neural network, generative adversarial network" as the generative model, and in Column 2, Lines 14-16 teaches "a dataset generator can generate a synthetic dataset for training the data model using a generative network of a generative adversarial network." Truong therefore fully teaches that the generative model includes a neural network. As for the "image, video, or sound" limitation, Kalchbrenner has been newly relied upon in the rejection of Claim 1 above and expressly teaches at Column 3, Lines 14-24 that "output example 152 can be a sequence of audio data, e.g., a waveform . ... As another example, the output example 152 can be an image . ... As another example, the output example 152 can be a video frame." The motivation to combine Kalchbrenner with Truong and Raman is to leverage the structure of output examples to achieve speed ups without degradation in quality, as discussed by Kalchbrenner. Hence the combination of Truong, Raman, and Kalchbrenner fully teaches these limitations. Therefore Applicant's remark cannot be considered persuasive.
In regard to Claims 4-7 and 13-16, they directly/indirectly depends on independent Claim 1, 19, 20 respectively. Applicant does not argue anything other than the independent claim 1, 19, 20. The limitations in those claims in conjunction with combination previously established as explained.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20180336455 A1 GENERATING OUTPUT EXAMPLES USING BIT BLOCKS
US 6941289 B2 Hybrid neural network generation system and method
US 9858524 B2 Generating natural language descriptions of images
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to YUJANG TSWEI whose telephone number is (571)272-6669. The examiner can normally be reached 8:30am-5:30pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kent Chang can be reached on (571)272-7667. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/YuJang Tswei/Primary Examiner, Art Unit 2614
